TWI304957B - Demand forecast systems and methods, and computer readable medium thereof - Google Patents

Description

1304957 IX. Description of the invention: [Technical field to which the invention pertains] The present invention relates to supply chain management, and a demand forecasting system and method for an adjustment mechanism = η - species with a gray-scale prediction, [prior art] σ In the supply of goods, the supply chain includes the procurement of materials, the conversion of materials into half products, and the distribution of finished products to customers. In other words, not every distribution capability is limited by the commitment of the capital but it will not be met. Among them, _4=! is enough to satisfy '- some customers may require the card to serve part of the supply, and others! 0 5 ^5 mention. ,...,, (4) The important pre-θ of the product manufacturer and service provider of the manufacturing and distribution, the customer transmits the requirements of the product including the specific quantity and the special period or = 4, the applicant should read the data Received requirements to measure and count enough internal = external manufacturing schedules, and are configured in advance to manufacture product capacity to meet each customer. The households will receive the order of the demand for the clearing, and the supplier will start to manufacture the product. The second person will solve the problem according to the county's demand for investment. _ Equipment and materials are in demand. The measurement cannot be effectively approached to the actual order. The supplier will suffer huge losses. "Bai Zhidi' lock sales miscellaneous data - domain (four) curry (four) or time series (7) coffee series) = type prediction: the regression model assumes that the variable __ is linear. However, due to the ambiguous market, θ Λ 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少 很少Time series models require a large number of experimental reference values (usually more than % pen data or more). In addition, the results of regression or time series models are usually simple measurements of the lack of impurities. Therefore, for the diversity of market life with limited remuneration and limited historical reward data

05O3-A3O865TWF 1304957 These forecasts are inconsistent for product type companies, such as upstream and downstream plants in the semiconductor industry. SUMMARY OF THE INVENTION In view of the above, the present invention provides a demand prediction system and method. * u According to the requirements of the present invention, the demand measurement system includes a billing group and an adjustment module. The planning module provides a predictive error interval mechanism to generate an error interval based on historical prediction data and a gray gamut. The adjustment module is based on the deletion error interval adjustment—the target prediction value of the target cut area. ~ Leaf stroke module is more normalized (N. Cong lize) historical prediction dragon, historical data corresponding to _ special = ten stroke interval, or - continuous calculation _. If the historical pre-counting to the specific counter-measurement (4) module is based on the actual _ actual order to normalize the historical forecast data. The juice drawing module uses the two predicted values generated from the historical prediction data and the gray prediction model to generate a prediction error interval. The grey prediction model is based on the theory of grey prediction theory. The module is further adjusted according to the predicted error interval and the distance of a predetermined reference hard magnetic (10) butterfly. «The requirements of this hair tender _ method, first of all, the shame data and a gray = polar generation - compensation error interval. Then, based on the prediction error interval adjustment - the target prediction value of the target d interval. θ A historical bead material is more standardized. Historical forecast data correspond to – specific plans, or capital rules. If the historical foot is corresponding to the specific meter, the historical pre-calling is normalized according to the actual order of the special section. The value is the prediction of the gamma history and the gray pre-sense. The fire color change type is based on the gray prediction theory (3) model. The target prediction value is further adjusted according to the _ error interval and the distance of the established reference system. The above-mentioned green phase transparent touch code of the present invention is recorded in the real needle. When the program is

0503-A30865TWF 6 1304957 The manned and executed day of the machine is turned into a rib to implement the apparatus of the present invention. The above-described objects, features and advantages of the present invention will become more apparent from the following detailed description. [Embodiment] FIG. 1 is a schematic diagram showing the architecture of a demand prediction system according to an embodiment of the present invention. The prediction system 1 (10) includes a planning module 11 () and an adjustment module (9). The calculation group 110|d- deletes the error interval mechanism based on the historical reward data (10) and the listening (((4) the prediction handle 111 generates the i measurement error interval. Laiyi nQ can first normalize the required historical prediction data 130. The purpose of formalization is to make the historical forecast data 130 fresh at a specific baseline (-). The required historical data (10) can be applied to the -planning interval, or a continuous planning interval. For example, A plurality of data sets corresponding to the planned interval in January, or a plurality of data sets corresponding to the planned intervals of the month, February, March, and April, respectively, may be used to generate a prediction error interval through the mechanism. The historical forecast data 13〇 corresponds to a continuous planning interval, and the planning module 110 normalizes the historical forecast data according to the average actual order of the planned intervals, the actual order of the subsequent planned interval, or other values. The historical forecast data 130 corresponds to a specific plan interval, and the plan module 11() normalizes the historical forecast data 13 according to the actual order of the specific plan interval. Table 1 shows the historical forecast data 130 Examples of raw materials Table 1 Juice interval 2003/January 2003/2 month historical forecast data (1) 1200 1300 Historical forecast data (2) 1100 1400 Historical forecast data (3) 1000 800 Historical forecast data (4) 1200 700 Historical forecast data (5) 1300 1000 Historical forecast data (6) 1050 Bu 900 Actual order 1100 900 0503-A30865TWF 7 1304957 The historical forecast data 130 of the planned interval in January includes 12 (10), 11〇〇, 1〇〇〇, 12〇 〇, 13〇〇, and 1050. The historical forecast data for the planned interval in February 13 includes 13〇〇, 14〇〇, 8〇〇, 7〇〇, 1000, and 900. Historical forecast data for January and February 13〇 can be normalized with the corresponding actual orders Π 00 and 900. The historical forecast data 13 is being displayed in Table 2. Table 2 Planning interval ~~ 2003/January 2003/2 _ Historical forecast data (1) ~ 1.0909 1.4444 Historical forecast data (2) 1.0000 1.5556 _ Historical forecast data (3) 0.9091 0.8889 - Historical forecast data (4) 1.0909 0.7778 Historical forecast data (5) 1.1818 1.1111 _ Historical forecast data (6) 0.9545 1.0000 Actual order, 1.0000 1.0 The 000 grey prediction model 111 can be a GMGJ model based on the grey prediction theory. The grey prediction is based on the theory put forward by Dr. Deng in 1982, focusing on hesitant and limited case samples. The grey prediction model only needs 4 samples to predict nonlinear behavior. It is worth noting that the accuracy of the prediction can be increased when using more samples. The normalized historical prediction data 130 may be an input of the grey prediction model m, and two predicted values may be generated by the _ grey prediction model ill. The planning module 110 uses the two predicted values to generate a prediction error interval. The adjustment module 120 receives the target predicted value 14〇 of a target planning interval, and adjusts the target predicted value according to the predicted error interval mechanism and/or the distance between the predicted error interval and a predetermined reference value. It is worth noting that the reference value can be the normal value of the actual order, such as 1〇〇〇〇%. The predicted result 15 〇 (adjusted target predicted value) generated by the adjustment module 120 can be used to predict demand/sales. It is worth noting that the target planning interval can be any planning interval. For example, to remove the effects of seasonal changes, this specific and target planning interval can be the same month in different years. Figure 2 is a flow chart showing a demand forecasting method in accordance with an embodiment of the present invention. First, as in step S210, the historical prediction data is normalized. As mentioned earlier, historical forecasting resources 8

0503-A30865TWF 1304957 It can be used to - specific _, or _ continuous meter room. If historical forecasting = corresponds to - consecutive footage, historical forecasting data can be normalized based on average consecutive orders, final-calculations, actual orders, or other values for these planned intervals. If the historical forecast data corresponds to - specific meter, the historical side data can be normalized according to the actual order in the specific planning area. As in step S22, the prediction data is generated based on the historical prediction data and the gray prediction model. In this example, the GM based on the grey prediction theory is used (1 mountain model. First, normalize dz(1),

The historical prediction data is input into the gray prediction model (1Γ + pull (1) = 预测 mid-predicted parameters (a approximation. It can be used to count two side values. The remuneration can produce a prediction error interval in the county. In the example of Table 2, use fine/ The six historical predictions of Haoyue, the two grey prediction parameters are: \ fa

Vi = 2 \ 2=2 5 χ ikr ι=2 6 Λ 2Σ-ω U=2 ) a 0.0171 and

(i) b Σζ(ι) U=2 6Σ 1 = 2 5 χ Σ (zd)2) - Σ Ζ(ΐ) U = 2 (i) =0.9654 where (6) is the historical data (, (1) I, the original sequence (Series), Ζ(1) is the AG〇(Accumulated Generating 〇perati〇n) sequence, and z(1) is the average sequence. It is worth noting that the 'grey prediction theory is a well-known technique, the details of which are omitted here. After that, the predicted value (Z(〇)(7)) is: 1 + 0.5a + 〇.5a. 1.0810 - 108.10% and

0503-A30865TWF 9 1304957 χ(6)(7)={Bu(6)(1)-兰 χ exp(_ 6a)} + 7 _ Fork (1)(7)=ι. 8. 9 = 108·09% Similarly, using the six historical projections of 2003/February, the two predicted values were 61.66% and 77.49%, respectively. Figure 3 is a schematic diagram showing the prediction error interval for the corresponding planning interval 2 (8) 3/1 month and 2003/2 month. As shown in Figure 3, the forecast error interval for 2003/January is between 108.10% and 108.09%, and the distance D1 to the reference value (100%) is 8.9%. The 2003/February forecast error interval fez:2 is between 77.49% and 61.66%' and the distance to the baseline value (100%) is 22.51%. Thereafter, in step S230, the target predicted value of a target plan interval is adjusted according to the distance between the predicted error interval and the predicted error interval and the reference value (e.g., 100%). Among them, the benchmark value indicates the market trend of the forecasted commodity. If the prediction error interval is south than the reference value, adjust the target prediction value according to the following equation: ATFV = TFV X (l - (si x FEl)) x (l - (s2 x D)) (work) If the prediction error interval is lower than The reference value is adjusted according to the following equation: ATFV = TFV X (1 + (si X FEl)) χ (ι + (s2 χ (7) where (4) is the target predicted value, which is the adjusted target predicted value, The prediction error interval, 〇 is the distance between the prediction error interval and the reference value, and the 釭 and illusion are adjustable constants.

It is worth noting that & corrections can be dynamically adjusted based on the type of comfort, production, customer conditions, historical forecasting accuracy, market trends and other factors. As mentioned above, the tears can be sighed. If the historical data corresponds to the continuous cap m, such as thin λ month ~ · tearing the moon, the target meter can be fine η month. In this example, seasonal motorized images can be removed in steps. In the example of Table 2, the 6-day lunar test of 2 months: the input value (10) reading, and the 6 historical prediction data of 2003/February can be used to predict the target forecast value of the cut interval class/month. . In addition, the target of the 20th month "Ln2〇〇4/2 month target 2004/February target can be adjusted using the public (2) with the formula (1), and the value can improve the county secret, the resident ==^ Target forecast

0503-A30865TWF 10 1304957 Body. Shows the storage media history of the computer program that stores the coffee test method _=:== The code contains at least _411, _ based on the prediction error ugly and the handle 412, used for the target prediction value of the interval, "=曰&amp ; distance adjustment between the quasi-values - historical prediction data of the target plan. 61 σ # ^ sensitivity includes code (fine *), rib normalization

- Figure 5 is - not intended to show the predicted trend in accordance with an embodiment of the present invention. For example, the 5th = = error interval 舆 舆 forecast trend, away) are two = Γ Γ wheel, her TM light lang ^ = H source, transfer and inventory system transfer. Including the wheel 'materials, TM__ state package can read the frequency, 3, ^, hard disk, give any of its _ _ _ computer becomes used in the ' ^ type code by the machine, such as computer manned and executed This machine has a frequency of 1 fx Ming device. The green device and the device of the present invention can also be transmitted in a code type, a process wire or an electric iron, an optical fiber, or any transmission type, and the second and the second are: when the computer receives, loads, and executes. 'This machine has been used to participate in the program, and the butterfly has been sprayed on. The present invention has been disclosed in the preferred embodiment as above, but it is not intended to limit the invention, and the person skilled in the art does not deviate from the present invention. Within the spirit and scope of the invention, when some changes can be made and /s, therefore, the mine of the present invention should be regarded as the special fiber riding industry;

0503-A30865TWF 11 1304957 Schematic description of the schema] = a butterfly system display according to the embodiment of the present invention - a flow chart showing a need for a detailed diagram according to an embodiment of the present invention - not intended to display a corresponding planning interval 2 〇〇 3 / 1 Error interval. Forecasting Mistakes in Months and 2003/February 4 Figure 4 is a diagram showing the storage of computer programs that provide a method for predicting demand in the library. Figure 5 is a schematic view showing the forecasting trend according to an embodiment of the present invention. [Major component symbol description] 100~ demand forecasting system; 111~ gray forecasting model; 130~ historical forecasting data; 150~ forecasting result; FEI, FEI1, FEI2 110~ planning module; 120~ adjusting module; 140~target prediction Value; S210, S220, S230~ operation steps; ~ prediction error interval; ΰ, D1, D2~distance; 400~ computer system; 410~ storage medium; 411, 412~ code; τ~ prediction trend.

0503-A30865TWF 12

Claims (1)

^I3'04957 Application for the underlined part of No. 94125907 ^, the scope of patent application: 1 · A demand forecasting system 9S year?. Package S' Revision date: 95.7.24 曰修(3⁄4)本本-plan module' Generating two values based on the historical data and the _ gray reward model, and generating a prediction error interval based on the two predicted values; and - adjusting the weight group for using the side error interval and/or the error interval and the value The distance adjusts a target predicted value of a target plan interval. ^ / 2. The system of demand for rewards as described in the patent application item 1, wherein the accounting group more formalizes the historical forecast data. 3. The demand forecast (10) as described in the patent application item 1, wherein the New Zealand forecast data corresponds to a specific plan interval, and the plan module normalizes the historical forecast data according to the actual order of the specific plan interval. . 4. The demand forecasting system described in claim 1 of the patent scope, wherein the historical forecast data corresponds to a continuous planning interval. 5. The demand forecasting system as described in claim 1, wherein the gray predictive model is a GM (1, 1) model based on grey prediction theory. 6. The demand forecasting system of claim 1, wherein when the predicted error interval is higher than the reference value, the adjustment module adjusts the target predicted value according to the following equation: ATFV = TFV X (l - (si x FBl)) x (l - (s2 xd)) where TFV is the target prediction value, which is the adjusted target prediction value, Qiugong is the prediction error interval, and 〇 is the prediction error interval and the reference The distance between the values, and 31 and the illusion are adjustable constants. 7. The demand forecasting system of claim 1, wherein when the predicted error interval is lower than the reference value, the adjustment module adjusts the target predicted value according to the following equation: ATFV = TFV X (l + (si x FEl)) x (l + (s2 xd)) where TiV is the target predicted value, ATFV is the adjusted target predicted value, FEI is the predicted error interval, and 〇 is the predicted error interval and The distance between the reference values, and S1 and 52 are 13 0503-A30865TWFl/Yianhou "1304957 Amendment date: 95.7.24 The patent scope of the 9412907 line portion is corrected by the adjustable constant. 8_—a method for predicting demand, comprising the following steps: generating two measured values according to a history, a m-material, and a gray-reward model, and generating a prediction error interval according to the two predicted values; and/or the error between the predictions based on the prediction The distance between the zone_-reference values adjusts a target predicted value of a target plan interval. 9. If the demand forecasting method described in item 8 of the patent scope is claimed, it further includes normalizing the historical forecast data. 10_If the application for the item of interest in the eighth item is applied, the method further includes normalizing the county historical data according to the __ specific planning interval: the historical forecast data corresponding to the specific planning interval. 11. The demand forecasting method of claim 8, wherein the historical forecast data corresponds to a continuous plan interval. 12. The demand forecasting method described in claim 8 of the patent scope, which makes the gray prediction model a GM (1, 1) model based on the grey prediction theory. 13. The demand forecasting method described in claim 8 of the patent application, further includes adjusting the target predicted value according to the following equation when the predicted error interval is higher than the reference value: ATFV = TFV X (l - (si X -FBX)) x (l ~ (s2 xd)) where ' rFV is the target predicted value' ATPV is the adjusted target predicted value, FS is the predicted error interval, and 〇 is the predicted error interval and the reference value The distance between them, and si and S2 are adjustable constants. 14. The demand forecasting method described in claim 8 further includes adjusting the target predicted value according to the following equation when the predicted error interval is lower than the reference value: ATFV = TFV X (l + (si X x (l + (s2 XD)) where ΓΡΥ is the target predicted value 'ArFV is the adjusted target predicted value, FBI is the predicted error interval, and D is the distance between the predicted error interval and the reference value, and Red and illusion is 0503-A30865TWFl/Yianhou 14 '1304957 No. 94125907, the scope of the patent application is revised. The date of this amendment is: 9.5.7.24 Adjustable constants 15. Computer-readable media, storage for a computer program Executing, causing a computer to perform a demand prediction method, the method comprising the steps of: generating two predicted values according to historical prediction data and a gray prediction model, and generating a prediction error interval according to the two predicted values; and calculating the error interval according to the prediction And/or the distance between the predicted error interval and a reference value adjusts a target predicted value of a target planned interval. 0503-A30865TWFl/Yianhou 15